Signals transmitted in voice communications include a sound signal and a soundless signal. For the purpose of communication, voice signals generated by talking and uttering are defined as a sound signal. A signal generated in the gap between the generally discontinuous uttering is defined as a soundless signal. The soundless signal includes various background noise signals, such as white a noise signal, a background noisy signal and a silence signal and the like. The sound signal is a carrier of communication contents and is referred to as a useful signal. Thus, the voice signal may be divided into a useful signal and a background noise signal.
In the prior art, a Code-Excited Linear Prediction (CELP) model is used to extract core layer characteristic parameters of the background noise signal, and the characteristic parameters or the higher band background noise signal are not extracted. Thus, during the encoding and decoding, only the core layer characteristic parameters are used to encode/decode the background noise signal, while the higher band background noise signal is not encode/decoded. The core layer characteristic parameters include only a spectrum parameter and an energy parameter, which means the characteristic parameters used for encoding-decoding are not enough. As a result, a reconstructed background noise signal obtained via the encoding-decoding processing is not accurate enough, which makes the encoding and decoding of the background noise signal of bad quality.